March 2013 - Institute of Physics and Engineering in Medicine

STEREOPHOTOGRAMMETRY FEATUREcan calculate distance measurements, surface area andvolume with various degrees of accuracy. 7 Approximateestimation of breast volume may be made from a modelof half an ellipsoid, based on manual linearmeasurements. Radiological techniques such asmagnetic resonance imaging, x-ray mammography andultrasound have been assessed. 8 These techniques areroutinely used in the diagnosis of breast cancer althoughthey are expensive, may involve the use of ionisingradiation and are often not practical for routineassessment of morphology. Conventionalphotogrammetry is routinely used as a tool for patientcommunication and documentation of post-surgicaloutcome; however, as a 2D projection of a 3D object, itprovides limited morphological information and servesprimarily as a pictorial record. The breast form has beenclearly defined in the literature in three dimensions, twoof which refer to the breast footprint on the chest wall(defined by upper, lower, lateral and medial borders)whilst the third dimension defines the volume,projection and overall shape of the breast. 9 Each of theseproperties can be accurately captured using 3D surfaceimaging technology in a way which linear measurementand conventional 2D photography do not allow.Compared to traditional anthropometry,stereophotogrammetry has several advantagesincluding non-contact assessment and low patientcontact time and eliminates random measurementerrors due to soft tissue deformation as measurementsare made on the images rather than directly on thesubjects. 10 Measurements obtained from 3D surfaceimaging have the potential to be much moreencompassing and are potentially more accurate andprecise. In addition, digital image analysis may enablepreviously unattainable measurements of breastparameters, such as surface area, volume andasymmetry, to be easily quantified and allow forquantitative evaluation in a longitudinal setting. Finally,stereophotogrammetry also has a role in establishing astandardised system for breast analysis, necessary toaccurately assess surgical outcomes in a reproduciblemanner. Tepper et al. introduced a framework termedmammometrics, defined as ‘the establishment of fixedplanes and points to perform objective breastmeasurements’. 11 These planes and points are defined infigure 2. The use of mammometric planes and points inconjunction with stereophotogrammetry data wouldallow changes in shape, contour, volume distributionand position of the breast on the chest wall to bequantitatively determined.Purpose of workReliability and stability of image capture, accuracy andrepeatability of anatomical landmark placement andlimitations of the stereophotogrammetry system mustbe investigated before the system can become anaccepted tool for clinical research and be integrated intothe hospital environment. This study assessed acommercially available stereophotogrammetry system:Di3D (Di3D.com, Glasgow, UK). The technicalperformance of this system was previously investigatedfor geometric accuracy, reproducibility and maximumfield of view in capturing images of the face using 50mm lenses. 12 We adapted the Di3D system using wideanglezoom lenses for acquisition of the torso for use inbreast assessment. The overall aim of this study was toassess the accuracy, reproducibility and stability of ahigh-resolution stereophotogrammetry system to imagea female mannequin torso. A key objective was todetermine if measurements made on the textured virtualsurface computed from stereophotogrammetry areequivalent to traditional anthropometric measurementsmade directly on the mannequin. This approach wasdeveloped to make objective measurements of thefemale breast, and potential applications are discussed.MethodThree-dimensional surfaces were acquired using theDi3D system which consisted of two camera pods, eachpod containing a pair of 10.1 megapixel colour digitalCanon EOS 1000D and a synchronised SunPak 383Autoflash. The system captures two stereo pairs ofimages and reconstructs the 3D surface by matchingdata from each set of images (stored in JPEG format).The textured surface data consists of two components:the 3D mesh, made up of four-sided polygons, and theoverlying high-resolution texture.The system was used to image a female mannequintorso made of expanded polycarbonate. Red paint wasapplied in a speckle pattern to the surface of themannequin to add texture as stereophotogrammetryrequires fine image detail at a pixel level to performaccurate image registration between the pairs ofphotographs. There should be sufficient skin detail inhuman subjects, for example freckles, skin tonevariation and moles, to accurately compute the 3Dsurface. A series of 17 landmarks were marked with afine point pencil, of point diameter approximately 1.0mm, distributed across the torso. The anatomicallocations of these landmarks are illustrated on a 3Dimage of the mannequin in figure 3. The camera systemwas calibrated in accordance with manufacturer’sinstructions. Wide-angle zoom lenses (Tamron 18–55mm, tamron.com, Saitama, Japan) replaced the standard50 mm lenses supplied with the system to ensure greatercoverage of the mannequin. The mannequin waspositioned facing the centre of the camera configurationat a distance of 90 cm and imaged using system settingsrecommended by the manufacturer [shutter speed 50ms, ISO speed 100, flash brightness ¼] and a nominalzoom setting of 24 mm. The surface was reconstructedusing Di3Dcapture, V5.2.5. Several experiments werecarried out to assess both the accuracy and repeatabilityof the image capture, with emphasis on the validity ofthe system for taking measurements of the breast.System repeatability and accuracyThe purpose of this experiment was to determinesystem repeatability and accuracy of softwaremeasurements compared to manual physicalmeasurements on the mannequin and to measure interraterreliability. The mannequin surface was captured 10times at short regular intervals and the textured surfacesreconstructed. A mean surface was also generated. ‰FIGURE 1.[TOP]3D surfaceimage of breastreconstructionshowing (left)high-resolutiondigitalphotograph ofthe skin(texture),(centre) 3Dsurfacerendering of(right) the 3Dpolygon mesh▼FIGURE 2.[CENTRE LEFT]Mammometricsframework forobjective breastassessment(used withpermission fromDr Nolan Karp,from Tepper etal. 11 )▼FIGURE 3.[CENTRERIGHT]Anatomicallandmarks usedfor thevalidation ofmeasurementson mannequinsurface▼FIGURE 4.[BOTTOM]Colour-codedsurfacedifference mapscomparing 10repeatedstereophotogrammetryacquisitions tothe meansurface (colourscale inmillimetres)▼SCOPE | MARCH 2013 | 21